Tag: ggplot2

Categories
Uncategorized

Video Game Genres by Year: 1980-2016

Video games have been a hobby since childhood so it’s fun to look back at the history of video games. I found a cool dataset that includes all video games with sales of 100,000 or more from 1980-2016. It’s necessarily incomplete since it was scraped from an existing website; this is especially true for the early days of gaming (e.g, it has only 98 Nintendo Entertainment System (NES) games out of the library of ~716). It also misses a lot of the download-only market for PC and has no mobile games. Nonetheless, it’s still an interesting snapshot of video game history. For my part, I’m interested in tracking the popularity of video game genres from 1980-2016. That is, the type of game (e.g., platformer, shooter, role-playing). Starting out, I’d presume that platformers (e.g., Super Mario) would have been most popular back in the NES and Super Nintendo / Sega Genesis era (the halcyon, nostalgia-ridden days of my youth). But let’s see what turns up!

First thing to note is that there is really not nearly as much data here prior to ~1995 (shortly after the Playstation 1 comes out). I ran a simple stacked bar chart first to look at this.

ggplot(newdata, aes(x = Year2, fill = Genre)) +
  geom_bar(position = "stack") +
  scale_fill_ptol() +
  scale_x_continuous(limits = c(1980,2016), breaks = seq(1980, 2016, 5)) +
  labs(x = "", y = "") +
  ggtitle("Video Games by Year and Genre: 1980-2016")

The problem with this chart is that I can’t make any sense of how popular different genres are because of the large differences in counts. Thus, it makes more sense to look at proportions. I decided to go with a stacked density plot this time, and I’m pleased with the result.

ggplot(newdata, aes(x = Year2, group = Genre, fill = Genre)) +
  geom_density(adjust=1.5, position="fill") +
  scale_fill_ptol() +
  scale_x_continuous(limits = c(1980,2016), breaks = seq(1980, 2016, 5)) +
  scale_y_continuous(limits = c(0,1), breaks = seq(0,1,.10)) +
  labs(x = "", y = "") +
  ggtitle("Video Games by Year and Genre: 1980-2016")

To interpret the results, it might help to know the broad eras of video game consoles:

  • (1976–1983): Atari 2600
  • (1983–1987): NES.
  • (1987–1993): SNES/Genesis.
  • (1993–1998): PS1/N64/Saturn.
  • (1998–2005): Dreamcast/PS2/Gamecube/Xbox.
  • (2005–2012): PS3/360/Wii.
  • (2012-Present): PS4/XB1/WiiU.

First, remember that these are only the games that sold more than 100,000 copies, so it’s more an index of popularity than actual games published. With this in mind, a few interesting trends emerge.

  • Action games are most popular in the Atari era (e.g., Frogger), and have picked up again substantially in the present-day era of console gaming.
  • Adventure games have slowly increased in popularity over time.
  • Fighting games hit their peak in the early 90s. Games like Mortal Kombat, Street Fighter, and King of Fighters are popular here.
  • As expected, platformers are huge during the NES and early SNES era. The popularity of Super Mario Bros. spawned a lot of clones.
  • The big spike and then decline of puzzle games is probably an artifact of the data. Puzzle games are probably even more popular now with mobile phones, but those games aren’t included in the dataset. I expect these games were much easier to program, back in the day, which accounts for the early spike (e.g., Tetris).
  • Racing games get popular in the 1995-2005 era. The popularity of game series like Mario Kart spring to mind here.
  • Roleplaying games don’t emerge as top sellers until the NES era, and stay pretty steady after that. I expected to see spike of roleplaying games in the PS1 era, considered by many to be the golden age of JRPGs. But probably doesn’t get picked up in this dataset because many of them are cult classics, and did not sell a lot in the US.
  • Strategy games don’t emerge as top sellers until the SNES era, but remain pretty steady after that.

Overall, very cool data and fun to look through. As usual, data and code available on the blog’s OSF site.

Categories
Uncategorized

Lollipop Plot: Meta-analysis of Gender Differences in Sexual Behaviour

When I went to the APS conference in San Francisco last year, I got to hear Janet Hyde talk about the gender similarities hypothesis. Broadly, she argues that most gender differences (i.e., men vs. women) in psychological variables tend to be small in size. She used meta-analysis — statistically summarizing the results of lots of published research — as a method of testing her hypothesis. I thought it was fascinating stuff and a great talk, so I wanted to incorporate some of her research into intro psychology. Since I’ve been sprucing up the intro psych section on sex and gender, I thought it would be interesting to include Petersen and Hyde’s (2010) meta-analysis on gender differences in sexual attitudes and behaviour. Broadly, it found that there are a few large sex differences (e.g., men watch more pornography and masturbate more than women), but consistent with her general argument, most differences were very small in size (i.e., smaller than d = .20) and men and women are more similar than different. However, the paper itself doesn’t present the results in a very PowerPoint friendly way, and certainly not intelligible to freshmen students:

Portion of Table 1 from Petersen & Hyde (2010)

So, with that in mind I decided to make a plot up to better visualize the data. The degree of uncertainty displayed by the 95% CI is important … but most in the class don’t actually have any statistics background. So I decided to teach them what “cohen’s d” was first, then create a graph that showed off those numbers. I also taught them that values less than .20 are essentially negligible, so I wanted to highlight that on the plot. I decided on a lollipop plot, which is a bit more appealing than a bar plot. 

library(ggplot2)
library(ggthemes)
library(ggExtra)

#Order the variables in Rank Order

mydata$name <- factor(mydata$name, levels = mydata$name[order(mydata$d)])

#Create the Plot

  ggplot(mydata, aes(x=name, y=d, label=d)) + 
    geom_point(stat='identity', fill="black", size=6)  +
    geom_segment(aes(y = 0, 
                     x = name, 
                     yend = d, 
                     xend = name), 
                 color = "black") +
    geom_text(color="white", size=2) +
    labs(title="Gender Differences in Sexual Behaviour", 
         subtitle = "Negative numbers = more common in women; positive numbers = more common in men",
         x = "", y = "cohen's d (effect size)") + 
    ylim(-1, 1) +
    theme_bw() +
    geom_hline(yintercept = 0, linetype = "solid", color = "grey") +
    geom_hline(yintercept = -.20, linetype = "dashed", color = "grey") +
    geom_hline(yintercept = .20, linetype = "dashed", color = "grey") +
        coord_flip()
Lollipop plot showing meta-analysis results

Overall, I think the plot turned out pretty clear and was a lot more appealing than just putting a table of numbers for a more dynamic presentation. The ordering from largest to smallest helps a lot to guide the eye, and the dotted grey lines show the range of “trivial” effect sizes. So there are some substantial gender differences, but the effect sizes are actually a LOT smaller than you’d expect! For context, a cohen’s d of ~.60 would look like this (using this site to generate this quickly):

So phrased another way, if you selected a random man and a random woman, there is ~67% chance that the man will watch more pornography than the woman (the “common language effect size“). That’s not insubstantial, but also probably a lot smaller than a lot of people would think! Many of the effect sizes are vanishingly small. For a Cohen’s d of 0.20, there’s only a 55% chance that a randomly selected man will be more likely to engage in the behavior than a randomly selected woman. For that matter, men and women are actually more similar than they are different on all of these variables, if you consider the percent overlap among the distributions. Which broadly, is what Dr. Hyde’s point has been all along.

Code and data available on my OSF page.

Categories
Uncategorized

Open Sex Role Inventory

Did you know about the Open Source Psychometrics project? It has more than two dozen personality tests that are all free to use with a creative commons license and posts large, open access datasets for their validation? Wow. What’s even stranger is that this site has no university affiliation, so far as I can tell and I can’t find any info on the site’s administrator. They’ve collected data from ~300,000 people here, these datasets are massive and an incredible resource, but are also virtually untouched by the academy. It’s downright criminal.

Case in point, the Open Sex Roles Inventory (OSRI). I started looking up data on Bem’s Sex Role inventory to help enrich the gender section of my intro psychology class, when I came across this measure. The open dataset had ~20,000 data points in it and I thought it’d be fun to play around with. I sent an email to the site administrator asking about how to score it and validation info. No email response, but the website was updated four days later with a TON of info and modifications to the scale to make a 20-item, more well-validated short form from the original 44 items. Seriously, who is this wonderful person? These data could easily be published in any personality journal they wanted, but they’re just free for all to use with no glory or publishing prestige. Searching online, I found a single publication using the OSRI, and it was a Ph.D. dissertation that used the data to test out a new method of measurement invariance, for which this particular data was mostly incidental (finding that the 44-item version of the OSRI didn’t have measurement invariance across heterosexual vs. non-heterosexual groups.

Anyway, back to the point. I wanted to have a visualization for my intro psychology class to clarify the distinction between sex (i.e., biological/physiological) and gender (i.e., socio-cultural). Specifically, I wanted to talk about Sandra Bem’s theory published way back in 1974 that gender roles are spectrum with at least two dimensions (masculinity vs. femininity). I’m aware of challenges to this distinction (e.g., that sex/gender are inseparable, and that sex is also a spectrum), but broadly won’t get into that here. In any event, the dataset I have only has sex measured as “Male, Female, Other,” so a categorical model will need to do for now. My goal was to show that there is a sizable minority of people beyond cis-gendered men and women. The dataset is pretty large (~16,000 after missing data was removed), so I decided to use geom_hex() with the viridis color scheme. This crushes my computer, and takes a long time to run but in the end, after a lot of obsessing over details, I am happy with the result. Special thanks to this post on Stack Overflow that helped me solve an graphical glitch that stumped me for a long time. 

ggplot(mydata2, aes(x = FemSF, y = MascSF)) +
  geom_hex(aes(colour = ..count..), bins = 22) + 
  theme_classic() +
  facet_grid(. ~ gender) +
  scale_fill_viridis() +
  geom_hline(yintercept = mean(mydata$MascSF)) +
  geom_vline(xintercept = mean(mydata$FemSF)) +
  labs(y = "Masculinity", x = "Femininity" ) +
  geom_text(x=4.25, y=4.25, label="Androgynous", col = "darkslategray2") + 
  geom_text(x=2.0, y=1.75, label="Undifferentiated", col = "darkslategray2") +
  geom_text(x=2.0, y=4.25, label="Masculine", col = "darkslategray2") +
  geom_text(x=4.25, y=1.75, label="Feminine", col = "darkslategray2")
Hexagonal Scatterplot for the OSRI

I like the way this graph came out. It shows that indeed, you can predict self-identified male or female status with the items on this questionnaire. But, it also shows that there’s a lot of variation, with folks that are androgynous (i.e., high in both) or undifferentiated (i.e., low in both) in for males and females. The “Other” gender category looks more similar to the female category than the male category, but a little more in the center overall. When I did some t-tests comparing men vs. women, femininity had a cohen’s d of 1.14 and masculinity had a cohen’s d of 0.82. Those are huge effect sizes! Much larger than the majority of sex differences in the literature.

Out of general interest, I decided to do a few more quick plots to break separate out the heterosexual vs. the non-heterosexual participants (I binned homosexual, bixsexual, asexual and other into one category for comparison in this analysis). Nothing fancy this time around, just some simple density plots: 

ggplot(mydata2, aes(x = Fem, fill = gender)) +
  geom_density(alpha = .4) +
  facet_grid(. ~ orientation.d) +
  scale_fill_manual(values = pal) +
  labs(x = "Femininity", fill ="")

ggplot(mydata2, aes(x = MascSF, fill = gender)) +
  geom_density(alpha = .4) +
  facet_grid(. ~ orientation.d) +
  scale_fill_manual(values = pal) +
  labs(x ="Masculinity", fill ="")
Density plots for femininity and masculinity, split by gender and sexual orientation

One thing that is pretty striking about these plots is that the femininity and masculinity scales predict identifying as a heterosexual man or woman REALLY well, with massive effect sizes. Folks identifying as “other” fell somewhere in between, though a little closer to the female profile. However, for the non-heterosexual folks it’s a different story. The sex difference is still there for femininity, but less pronounced, mostly accounted for by men being more feminine. For the masculinity scale though, there’s basically no discrimination at all! Men, women and non-binary folks can’t really be distinguished based on their masculinity scores. I suppose that’s consistent with the measurement invariance issue the this Ph.D. thesis picked up on. Not sure what it means in a theoretical sense, but it’s definitely interesting.

I could stop here, but I wanted to also do a quick factor analysis on the short form data, just to see if the factor structure looked good. Nothing fancy or in-depth, just an exploratory factor analysis extracting two factors, with principal axis factoring, and an oblique rotation. I also fussed around until I could make a nice plot of the factor loadings, using the ggrepel package. 

#Factor Analysis

mydata4 <- select(mydata2, 
       Q4, Q26, Q28, Q30, Q34, Q36, Q38, Q40, Q42, Q44,
Q1, Q3, Q5, Q7, Q9, Q17, Q29, Q35, Q39, Q43)

corMatsf <- cor(mydata4)

solutionsf <- fa(r = corMatsf, nfactors = 2, rotate = "oblimin", fm = "pa")

#Made a .csv file manually by copying into excel (can't figure out better solution)

print(solution$loadings, digits=2, cutoff=0)
loadings.sf <- read.csv("efa.loadingsSF2.csv")

#The plot

ggplot(loadings.sf, aes(x = PA1, y = PA2, label = item, col = FemMasc)) +
  geom_label_repel() +
  geom_point(size = 3) +
  scale_y_continuous(limits = c(-.80, .80), breaks = seq(-.80, .80, .10)) +
  scale_x_continuous(limits = c(-.80, .80), breaks = seq(-.80, .80, .10)) +
  geom_hline(yintercept = 0) +
  geom_vline(xintercept = 0) +
  scale_color_manual(values = c("hotpink", "royalblue")) +
  labs(x = "Factor 1 (Masculinity)", y = "Factor 2 (Femininity)", col = "") +
  guides(col=FALSE)
Plot of standardized factor loadings for OSRI

That looks pretty excellent, items are clearly distinguished by their factors and the factors are mostly uncorrelated. So the measurement seems solid. No wonder, given the amount of data that went into validating it! Overall, very cool dataset and measure. Might even use this myself some day in a published study…

If you want to take the test yourself, take a look here: https://openpsychometrics.org/tests/OSRI/

Data and syntax available on my OSF site. Data were used as part of a CC BY-NC-SA 4.0 license, with attribution the mysterious owner of https://openpsychometrics.org.

Categories
Uncategorized

Pathfinder Monster Database: AC vs Touch AC

Ok, one more visit to this pathfinder monster database before I’m on to a new dataset. This time, I wanted to take a look at the relationship between Armor Class (i.e., how hard a monster is to hit) and Challenge Rating (i.e., how tough the monster is, according to the game developers). There should be a pretty strong linear relationship between AC and CR.

However, the thing I’m really interested in is the relationship between Touch AC and CR. Because pathfinder is needlessly complicated, monsters have a separate AC for just touching them (or getting hit with certain kinds of attacks like rays or bullets). It’s been my experience that touch AC is always low, and doesn’t seem to change at all as the game progresses. Pragmatically speaking, this means that at a certain point touch attacks basically always hit. Let’s see if that’s the case. I think some run-of-the mill scatterplots will do here, but might as well make them fancy. Three packages needed: (a) ggplot2 for the basic plotting; (b) ggExtra to add density distributions to the plot; (c) cowplot, to stitch the two plots together at the end and (d) ggthemes, in case I want some custom color palettes. 

library(ggplot2)
library(ggExtra)
library(cowplot)
library(ggthemes)

Then, I can make the plots. Made the dots transparent with “alpha = 0.3” to deal with overplotting. Personally, I find it looks cleaner than geom_jitter(); it’s always low-key bothered me that the randomness added by geom_jitter() actually changes the positions of the raw data a little. Feels a little like lying with data to me, sometimes. With the ggmarginal command, I added in the density plots so we can see the overall distributions. Then at the end, I used cowplot to stitch them together side-by-side. Knowing this, I made sure both graphs had the same y-axis lengths to facilitate easier comparison, a key component of the small multiples approach.

p1 <- ggplot(mydata, aes(x = CR, y = AC)) +
  geom_point(size = 2, alpha = 0.3, color = "steelblue") +
  theme_classic() +
  geom_smooth(col = "royalblue", se = FALSE) +
  scale_y_continuous(breaks = seq(0, 70, 5)) +
  scale_x_continuous(breaks = seq(0, 40, 5)) 

p1m <- ggMarginal(p1, type = "density", fill = "steelblue")

p2 <- ggplot(mydata, aes(x = CR, y = AC_Touch)) +
  geom_point(size = 2, alpha = 0.3, color = "steelblue") +
  theme_classic() +
  geom_smooth(col = "royalblue", se = FALSE) +
  scale_y_continuous(limits = c(0, 70), breaks = seq(0, 70, 5)) +
  scale_x_continuous(breaks = seq(0, 40, 5)) +
  labs(y = "Touch AC")

p2m <- ggMarginal(p2, type = "density", fill = "steelblue")

totalplot2 <- cowplot::plot_grid(p1m, p2m, labels = "auto", ncol = 2)
  
cowplot::save_plot("scatter.pathfinder.png", 
totalplot2, 
base_aspect_ratio = 2)
Scatterplots of AC and CR

Ok, well there’s a clear (very strong) linear relationship between CR and AC (r = .91). That’s probably not surprising, given that the game designers probably wanted that to happen by design, to keep up with increases to attack bonuses that players get each level. However, touch AC is very weakly related to touch AC (r = .12). In fact, it’s really only mythic level monsters beyond CR 20 that see any increase in Touch AC at all! So for the vast majority of play, there’s no relationship between CR and Touch AC. I got to wondering then, what is the variance in touch AC actually attributed to? A good first guess would be a creature’s size:

ggplot(mydata, aes(x = CR, y = AC_Touch, col = Size)) +
  geom_point(size = 2, alpha = 0.3) +
  theme_classic() +
  geom_smooth(method = "lm", se = FALSE) +
  scale_y_continuous(limits = c(0, 70), breaks = seq(0, 70, 5)) +
  scale_x_continuous(breaks = seq(0, 40, 5)) +
  labs(y = "Touch AC") +
  scale_color_ptol()

Yup, looks like the majority of variation is due to that. So in summary, touch AC doesn’t really change as monster CR increases, and most of the variation is due to size. It’s a weird mechanic to add to the game, and easily abusable by additions like the gunslinger that always hit touch AC. Far as I can tell though, hasn’t been removed from the second version under playtest. Oh well, add it to the list of weird broken mechanics, I suppose.

Scatterplots of touch AC and CR, split by size

Syntax and data available on the blog’s OSF page.

Categories
Uncategorized

Pathfinder Monster Database Plots

I want to incorporate more R into my classes at Dalhousie. Problem is, I am a pretty bad R coder– I spent much of the past decade or so with SPSS and Mplus. But there’s lots of evidence that R is the future of science. I find that the best way to learn is project-based, so I’m going to start blogging on R code. I’m going to focus on topics that are inherently interesting to me, with a focus on data visualization. If I keep it fun, I’m more likely to stick with it.

So, to start I’m going to analyze data from the Pathfinder Monster Database, a comprehensive database of all 2812 monsters from Paizo’s tabletop roleplaying game, Pathfinder. I’ve played Pathfinder for years now and there are a lot of crunchy numbers in there. Probably why I like it so much! I’m going to look at the relationship between creature type two outcome variables (a) Armor Class (i.e., how hard the creature is to hit) and (b) Challenge rating (i.e., how tough the monster is overall). The goal is to see what creature type is “toughest” overall.

The data needed a little bit of cleaning (e.g., changing “Dragon” to “dragon” for some entries), but it was in good shape overall. I decided to try out ridge plots as the way to visualize the data, since I’ve never used them before. First thing to do is load the necessary libraries into R. 


library(ggplot2)
library(ggridges)
library(dplyr)
library(ggExtra)

Next, since I want the two plots to be in order from highest to lowest values of AC/CR, I need to use the next bit of code which requires dplyr. This creates two new variables I can use to re-order the y-axis with later. I also created a color palette of 13 random colors, since there are 13 creature types and I didn’t like the default ggplot2 colors here.

<h1>Order variables by AC</h1>
avg <- mydata %>%
group_by(Type) %&gt;%
summarise(mean = mean(AC))

ACorder &lt;- avg$Type[order(avg$mean)]
<h1>Order variables by CR</h1>
avg2 <- mydata %>%
group_by(Type) %&gt;%
summarise(mean2 = mean(CR))

CRorder &lt;- avg2$Type[order(avg2$mean2)]
<h1>Create color palette</h1>
pal &lt;- rainbow(13)

Ok, now I can create the two plots using the geom_density_ridges() function. This needs the ggridges package to function, as base ggplot2 can’t do this. 


ggplot(mydata, aes(x = CR, y = Type, fill = Type)) +
geom_density_ridges() +
theme_ridges() +
theme(legend.position = "none") +
scale_y_discrete(limits = CRorder) +
scale_x_continuous(limits = c(0,30), breaks = seq(0, 30, 5)) +
scale_fill_manual(values = pal) +
labs (y = "", x = "Challenge Rating")

ggplot(mydata, aes(x = AC, y = Type, fill = Type)) +
geom_density_ridges() +
theme_ridges() +
theme(legend.position = "none") +
scale_y_discrete(limits = CRorder) +
scale_x_continuous(limits = c(0,50), breaks = seq(0, 50, 5)) +
scale_fill_manual(values = pal) +
labs (y = "", x = "Armor Class")

So, the toughest monster types in Pathfinder are dragons, followed by outsiders. The weakest monster types are vermin and animals. The ranking of toughness by CR and AC are exactly the same, as it turns out. However, the distribution for oozes are way different than everything else: These creature types tend to be really easy to hit, but are still tough because of lots of other abilities and immunities. The positive skew in the distributions for CR are interesting, since it shows that there are generally a LOT more monsters under CR 10, which makes sense given that very few games get to such high levels.

I like ridge plots. They work a lot better than overlapping histograms when there are lots of groups and lots of cases. There was a bit of difficulty with numbers less than 1 for the CR plot (e.g., some CRs are 1/3). Without the “scale_x_continuous(limits = c(0,50)” function, the graph displayed values less than 0, which is outside the range of actual data. I believe that the graph is now bunching all the CRs that are less than 1 (~217 data points) as “0” on the graph above. Overall, a fun first attempt, and neat data to work with.

Datafile and syntax available on the blog’s OSF page.